French artificial intelligence lab Mistral unveiled its first line of reasoning models, branded under the name Magistral. With this new release, the company joins an increasingly competitive field, following similar recent releases from industry leaders like OpenAI and Google.
The Magistral suite comprises two models: Magistral Small and Magistral Medium. Magistral Small, with 24 billion parameters, is immediately available for public download on AI development platform Hugging Face under the permissive Apache 2.0 license. Parameters in AI models are the underlying components that determine their functionality and effectiveness.
Magistral Medium, larger and considerably more powerful, is currently available in preview through Mistral’s proprietary chatbot service Le Chat, via APIs directly offered by the company, and through selected third-party cloud providers. According to Mistral’s official announcement, these reasoning models are specially tuned to handle complex, multi-step logic processes with greater consistency and reliability. This makes Magistral especially suitable for tasks in mathematics, physics, structured calculations, decision-making trees, and rule-based logical systems.
Despite significant investment support, including backing from prominent venture capital firms like General Catalyst and over €1.1 billion (approximately $1.24 billion) raised since its inception in 2023, Mistral has struggled to keep pace in certain key areas compared to other top AI research organizations. Benchmarks released along with Magistral indicate that Mistral’s Medium model trails behind Google’s Gemini 2.5 Pro and Anthropic’s Claude Opus 4 in tasks assessing capabilities in physics, mathematics, and science. Furthermore, Magistral Medium also failed to beat Gemini 2.5 Pro in LiveCodeBench, a widely accepted benchmark for evaluating coding proficiency.
Aware of these limitations, Mistral emphasized other strengths in its announcement. Notably, according to company claims, the Magistral models deliver solutions at speeds up to ten times faster than competitor offerings available via the Le Chat chatbot platform. Additionally, Magistral supports numerous languages, from Italian and Arabic to Russian and Simplified Chinese, significantly broadening its global applicability.
Mistral’s recent product announcements reflect an earnest effort to diversify its AI product range beyond conventional models. Previously, the company launched a developer-focused tool called Mistral Code aimed at improving user coding experiences, as well as introducing enterprise-oriented solutions such as Le Chat Enterprise. Through Le Chat Enterprise, businesses can integrate AI-operated processes with widely used services including Gmail and Microsoft SharePoint.
In its announcement on Magistral, Mistral highlighted that these models are not only built for research and strategic planning tasks but also optimized for real-world operational challenges, such as risk assessment incorporating multiple variables, ensuring data-driven decisions, and logistics planning with complex constraints.
